This is an example of visuals that we can create visualizations for a specific location and buffer zone around the location where we have model outputs on.
Time Series Plots
We used the AllModel30.csv output for this analysis.
Metrics we use include:
- Number of EDT Accidents (Model Dependent Variable)
- Number of Waze Accidents
- Number of Waze Jam
- Number of Hazard On Shoulder (nHazardOnShoulder)
- Number of Hazard On Road (nHazardOnRoad)
- Number of Hazard Weather (nHazardWeather) - is it # of hazard or weather? According to Michelle, this should be one of the three categories of Waze event (Accidents, Jam, and Hazard/Weather). However, in the dataset, nHazardWeather is smaller than sum of the sub-categories (e.g., nHazardOnShoulder, nHazardOnRoad). This needs further investigation.
- Probability of NonCrash and Crash - real prob from RF model (needs further investigation/decision making) The probability has converted to based on the threshold ~ 0.2 using the built-in function in RF.
To do 8/23/2018:
- total/average/max (point/line) number of accidents/jams for every day (done)
- plot on top of each other for the NoEvents with a different color. Use other colors for a type of special events. 4/23 (4/16 easter), 5/21 (5/29 memorial day, 5/14 mother’s day), 7/02 weekends near holidays.
- Do both 3-miles and 1-mile. Removing zeros.
- Panels by month.
- add shadebox for event duration
- add function for visuals (done)
- number of accidents per gridcell per hour (done)
- boxplot of accidents over 42 grid cells. (done)
- look at hour aggregation
- heatmap for each hour
- smaller buffer zone, 0.25, 0.5, 1 mile radius
- update SpecialEventsExpand dataset (done)
- update the analysis and visuals (done)
- Easten shore beach - Bay Bridge area (7/4, 9/4 Labor Day, 3lanes both ways) for a typical holiday pattern
- follow up with Erika tomorrow (done)
- write one plot function in stead of multiple plot functions. (done)
Dave’s feedback 8/29/2018:
- try large buffer, e.g., 5 miles (Dan’s suggestion: work with individual lag/lons instead of grid cells)
- broader patterns - can we add baseball game,(Michelle’s comments: baseball games are usually hosted in urban areas with intercity activities and good public transportation, less parking. Erika - How far aways is the parking. Look that the parking in the buffer).
- Two days of week (Saturday, Wednesday)
- Extend the 23 hours through another 6 hours, change time window from 5 am to 5 am.
- Other types of metric, Hazards on the roads, car stopped on shoulders. Car stopped on shoulders.
- Other location: for exmaple, University football games.
- Can we try to make this a little more of hot spots, can we do this on a grid cell level.
HeatMap
5 mile buffer zone

Example Days in Presentation
Comparing all Matrics